I'm trying to get a consise configuration for setting up a FC Volume and LUN for a VMware ESX 4 datastore. I've reviewed TR-3749 but ultimately don't know what the best settings or practices should be. We're looking to leverage "thin provisioning" but not sure if this is the right approach or not. We have a fairly small team, so the same team who handles the virtualization environment will also manage the Netapp. Also does anyone know when the updated version of TR-3749 will be released. I know it should be any day now.
I'm a newbie, so any help is greatly appreciated. Thanks.
We're planning to use NFS and FC (already have the infrastructure), RDM LUNS for SnapDrive, SMVI for VM snapshots, RCU 3.0 and VSC
FC volume for the VM LUN - What's the ideal configuration
1. Is it a good idea to "thin" provision the Volume by setting the volume "space setting" to none vs volume (space guarantee)
2. If the Volume is configured with space guarantee, then the fractional space reservation should be set to 0%, right?
3. Dedupe, volume autogrow, and snapshot autodelete should be enabled?
4. Out of these five settings, what should be enabled: No access time updates, create unicode, convert unicode, no automatic snapshot copy.hide snapshot directory
5. snapshot reserve set to 0%?
LUN for VM datastores
1. If the Volume was set as "space setting" none, then should the LUN space reserved be disabled.
2. If the Volume was set as "space setting" guarantee, then should the LUN space reserved be disabled.
3. Should "enabled LUN volume to grow automatically" be always set no matter how we configured the volume.
4. For the LUN size, should it match the size of the Volume?
For NFS Volumes used for datastores
1. Space setting set to none?
2. Fractional space reservation set to 100%?
3. Dedupe enabled
4. Allow volume to grow automatically and delete snapshots automatically be enabled?
5. Out of these five settings, which ones should be enabled: no access time updates, create unicode, convert unicode, no automatic snapshot copy, hide snapshot directory
Thanks for getting back to me. I have a few additonal questions that I hope you can clarify on.
1.NFS - When using RCU 3.0 to create the Volume/Datastore, the create_ucode is set to ON.Does it matter if it’s set to ON or OFF?In the TR3749 v2, Fig. 67 on page 85, the picture shows that the create_ucode is set to OFF.
2.RDM LUNS – Should the settings on the Volume hosting the RDM LUNS be treated the same as a Volume hosting Datastore LUNS?i.e. all Volume settings should be the same, if we were creating a thin RDM LUN then it should reside in a thin Volume.Snapdrive for Windows will be used to create the RDM LUN.
3.RDM LUNS – When creating RDM LUNS is it best to store them all within a limited number of Volumes or is it better to do a one to one ratio, i.e. RDMs LUNS map to their own Volume.
4.Per the TR, it is recommended that a dedicated “vswap” datastore be created.If we had two FAS controllers in a cluster, would we still only need one dedicated “vswap” datastore?
5.We’re running two FAS controllers in a cluster (active/passive) and have FC and NFS datastores spread across both controllers.For the “OS pagefile” datastore, do we need to create only one FC datastore that we can use for VM machines both hosted on FC and NFS datastores?
Re: Consise best practices for FC Volume/Lun configuration for vmware ESX 4
1. I beleive you may reding into the document a bit. Unless implicitly stated, one should be able to proceed using default array configuraiton settings. Please understand it is easier to write a document where we cover what to change versus covering every option whcih we do not change.
2. use the default volume settings unless the provisioning tool, in your case SnapDrive, states to modify the settings.
3. This question is best answered with condieration around the goal of the implementation. In the physical realm many like a single LUN in a single FlexVol. With RDMs in VMware there are advantages and disadvantages to deploying in this manner.
4. vswap is a single datastore connected to by all of the nodes in a cluster. it is a central cluster resource.
5. Please do not deploy VMs accrross both nodes of the controller. As for pagefile datastores, one is recommended for every produciton datastore. See TR-3428 or 3749 for these details.